R programming: vectors

R programming
S3 vectors
vctrs
Published

September 2, 2024

Vectors in R

vctrs package

Objectives

Atomic vectors

In R parlance, vectors denote very general forms of sequences, that is objects that can be indexed using [[]], subseted/sliced using [], and combined using c(). We often confuse vectors and atomic vectors. Figure 1 from Advanced R by Wickham outlines that atomic vectors are special cases of vectors (just as lists).

Figure 1: Atomic vectors as vectors
Reading assignments

Read first Vectors in R for Data Science 1st editon, then Vectors in Advanced R programming.

In words, atomic vectors are homogeneous vectors where all items have the same type. This criterion is questionable, since defining the type of an object in R is not obvious. There is a type hierarchy, and objects may have several types. Nevertheless we may use typeof() to determine the storage mode of an object.

Basic atomic vectors

Basic atomic vectors are sequences of objects with the simplest storage modes.

Figure 2: Common atomic vectors

Null values

Question

Try to determine which items in a vector are NULL

x <- c(NA, 3, NA, 7, 13)
x == NA

Explain the output. Fix it.

Question

What happens when you combine (with c()) atomic vectors with different base types?

x <- c(1L:3L)
y <- letters[5:9]
z <- rep(c(TRUE, FALSE), 2)[1:3]
x ; y ; c(x,y) ;  c(x,z) ; c(y,z)

Attributes

Question

Attributes are metadata.

x <- as_date("2024-08-06") + 1:7
is_vector(x) ; is_atomic(x) ; class(x) ; typeof(x)
[1] TRUE
[1] TRUE
[1] "Date"
[1] "double"
$class
[1] "Date"
names(x) <- wday(x, label=T, abbr=F)

x
    mercredi        jeudi     vendredi       samedi     dimanche        lundi 
"2024-08-07" "2024-08-08" "2024-08-09" "2024-08-10" "2024-08-11" "2024-08-12" 
       mardi 
"2024-08-13" 
$class
[1] "Date"

$names
[1] "mercredi" "jeudi"    "vendredi" "samedi"   "dimanche" "lundi"    "mardi"   
x[["mercredi"]]
[1] "2024-08-07"
attr(x, "names")
[1] "mercredi" "jeudi"    "vendredi" "samedi"   "dimanche" "lundi"    "mardi"

Less basic atomic vectors

Question

What are raw vectors good for?

Question

What is the difference between POSIXlt and POSIXct? (Ask chatgpt)

Question

What does is.atomic() do?

Question

Is it possible to have an atomic vector with type/class POSIXct? POSIXlt? Are the answers of classand typeof always identical/consistent?

Question

Explain the following

x <- "A Man A Plan a Canal Panama"
y <-  rep("A Man A Plan a Canal Panama", 5)
is.character(y) ; obj_size(x) ; obj_size(y)
[1] TRUE
136 B
176 B
Question

Is an object of type factor a vector? an atomic vector?

Question (Exercise 20.4.6.2 from R for Data Science 1st Ed)

Carefully read the documentation of is.vector(). What does it actually test for? Why does is.atomic() not agree with the definition of atomic vectors above?

Factors, Dates, and Date-times are cases of Augmented vectors.

Recalling S3 classes

A basetype object with at least a class attribute.

Attribute Class is used to implement the S3 object oriented system.

In the next chunk x is a vector with basettype double but class Date. Each item in x is interpreted as the number of days ellapsed since the origin of time according to Unix 1970-01-01. It is printed accordingly.

x <- as.Date("2024-08-06") + 1:7

class(x) ; typeof(x)
[1] "Date"
[1] "double"
#| code-fold: false
x |> 
  unclass()|>
  str()
 num [1:7] 19942 19943 19944 19945 19946 ...

as.Date() is an example of generic function from base R.

as.Date
function (x, ...) 
UseMethod("as.Date")
<bytecode: 0x571a8b795978>
<environment: namespace:base>

The chunk above used method as.Date.character()

methods("as.Date")  # methods("as.Date")
[1] as.Date.character   as.Date.default     as.Date.factor     
[4] as.Date.numeric     as.Date.POSIXct     as.Date.POSIXlt    
[7] as.Date.vctrs_sclr* as.Date.vctrs_vctr*
see '?methods' for accessing help and source code
getS3method("as.Date", "character")  # as.Date.character

Examples

Factors have basetype integer and attribute factor and levels.

ctr_names <- factor(ISOcodes::ISO_3166_1$Name)

ctr_names |> 
  str()
 Factor w/ 249 levels "Afghanistan",..: 13 1 7 8 2 3 6 234 11 12 ...
class(ctr_names); str(attributes(ctr_names))
[1] "factor"
List of 2
 $ levels: chr [1:249] "Afghanistan" "Åland Islands" "Albania" "Algeria" ...
 $ class : chr "factor"
ctr_names |> 
  unclass()  |> 
  str()
 int [1:249] 13 1 7 8 2 3 6 234 11 12 ...
 - attr(*, "levels")= chr [1:249] "Afghanistan" "Åland Islands" "Albania" "Algeria" ...

Examples of important S3 classes

  • lm
  • kmeans
  • prcomp
  • hclust
Question

Explain

ctr_names |>
  str_to_upper() |>
  str()
 chr [1:249] "ARUBA" "AFGHANISTAN" "ANGOLA" "ANGUILLA" "ÅLAND ISLANDS" ...

History

Relevance: S3 generics

An S3 object behaves differently from its underlying base type when it is passed to a generic function.

What is a generic?

Question
  • What happens if an S3 object is passed to a generic?
  • What is method dispatch?
  • What kind of MRO (Method Resolution Order) is used?
  • How would you register a new method for a generic?
  • How do you define a generic?
  • Give examples of generics in base R.
  • Get the list of base R functions which are generics.
Question

Preserving attributes

S3 vectors as collections

In ?@lst-simple-loop, collection may denote a list, a vector, or any other iterable sequence you can encounter in R.

Vectors deserve special consideration.

Documentation

S3 vectors in Advanced R Programming

Desirable properties of vectors

Combining vectors using c()

vctrs and S3 vectors

Package vctrs makes the life of developpers who rely on S3 vectors easier.

Creating a new S3 vector class

In package nycflights13, in tibble flights, columns with names ending with dep_time or arr_time have basetype integer.

Loading required package: nycflights13
flights |> 
  select(ends_with('_time')) |>
  glimpse()
Rows: 336,776
Columns: 5
$ dep_time       <int> 517, 533, 542, 544, 554, 554, 555, 557, 557, 558, 558, …
$ sched_dep_time <int> 515, 529, 540, 545, 600, 558, 600, 600, 600, 600, 600, …
$ arr_time       <int> 830, 850, 923, 1004, 812, 740, 913, 709, 838, 753, 849,…
$ sched_arr_time <int> 819, 830, 850, 1022, 837, 728, 854, 723, 846, 745, 851,…
$ air_time       <dbl> 227, 227, 160, 183, 116, 150, 158, 53, 140, 138, 149, 1…

Nevertheless, these columns encode time information (hour, minute, second) in an unusual way. The last two digits represent minutes, the leading digits represent hours. In the sequel, we define an S3 vector class with basetype integer that will allow us to handle these columns in a transparent way. Desirable properties are

  • Readable display: 517 should be displayed as 5h17m
  • Some time arithmetics should be possible: we should be able either to add difftime or to compute the difference between dep_time and sched_dep_time
  • Some validation should be possible: 2517 is not a valid value for dep_time
  • Casting to datetime should be easy
  • Casting from datetime should be easy as well
  • …

We use the tools from article S3 vectors

Question

Create a new S3 vector class called weird_tm. Endow it with a constructor new_weird_tm(), an helper weird_tm(), a test is_weird_tm().

Question

Define a format() function for class weird_tm. Mind NAs.

Question

Casting and coercion

The next piece of code does not work

c(weird_tm(flights$dep_time[1:5]), flights$dep_time[1:5])

We need to define casting methods for generics vec_cast() and vec_ptype2() at least for casting to integer and character.

Question

Transform the tibble flights so that columns with name ending with _time (except air_time) have type weird_time. Is it still possible to filter rows with dep_time is a prescribed time interval.

We will use tools from vctrs to define differences between weird_tm objects.

Question

Double dispatch.

Question

Define the difference - operator for two vectors of class weird_tm. The result is expected to be an integer vector.

Further reading